Related work, side by side

Comparisons

Three head-to-head pages, each governed by the same rule: state the other side at its true strength, credit prior work first, declare the programme's own maturity without flattery, and name the measurement that would decide. No page here declares a winner, because the deciding evidence does not exist yet.

Embedded correction vs evaluation-based alignment

The field's standard build-then-examine approach against the proposal that correction must live inside the recursive loop. One has scale without a proof of sufficiency; the other has a measurement design without scale.

The Eden Protocol vs Constitutional AI

Both reject bolt-on rules. Constitutional AI trains values into behaviour and runs at frontier scale; Eden proposes entangling care with capability itself and is at pilot scale in one laboratory. Prior work credited first.

Independent vs institutional alignment research

Not a superiority claim: a structural comparison of which experiments are cheap from which position, what independence buys, and exactly what it costs.

Corrections and missed prior work are actively invited; the prior work page names the intake path. Every claim on these pages traces to the registers.

Sources: the prior work register; the open problem page; the evidence spine.